Application Of Precision Medicine For Cancer Treatment

Just as our fingerprint is unique to ourselves, cancers also hold unique genetic characteristics. Thus, we can only expect that different cancer patients will respond differently to drug treatments. Here’s where precision medicine comes into play. Just as a tailor takes down the measurements of its clients to craft a perfectly fitted suit, precision medicine ensures that the treatment is tailored to the patient based on their genetic information. With the growing presence of precision medicine, researchers from Georgia Institute of Technology conducted a study aimed to bridge the gap in finding the optimal cancer treatment. They developed a machine learning algorithm that predicts the treatment that will work best based on the patient’s genetic data. To do this, they first compiled genetic information of cancer patients from the National Cancer Institute's panel. Then, they narrowed down to look at each patient’s gene expression (used to classify cancer tumours) and drug sensitivity profiles (used to see the effectiveness of the drugs on the patients). They found that there are similarities between the patient’s gene expression and their sensitivity to the drugs. This process of analysis is also known as training the data. Just as kids are taught math, the algorithm is “taught” to learn these patterns and holds it in its memory. When it is used to predict new sets of data, the algorithm simply works like a machine: First, we input the information that we do know – the genetic information of the patient.

The machine takes this information and finds similarities with the genetic information of previous cancer patients. It then analyses those cases to see what treatments worked best. The process ends with an output – a prediction for the treatment that will have the most favorable response for the current patient’s condition. Some may ask, “What’s the need for all this work? Can’t we just look to ongoing cancer research to find the cures?” The thing is, the current state of cancer research cannot properly explain the causes of all the different types and forms of cancer genes and we don’t know when those specific cures will be discovered. So, why don’t we just look right in front of us and use all the data we already have? Instead of waiting for those causes to be discovered, we are able to battle cancer by looking at correlations instead of causes. Instead of being frustrated not knowing how the tumor formed, we can be relieved to know that a treatment worked for someone with similar genetic information. Using this machine learning algorithm is effective to determine optimal treatments because it breaks past the boundaries undiscovered causes of all cancer types. People may expect that the best way to find the cure for one’s cancer would be to narrow down the search to the past cases of that specific type of cancer and see what cures worked best.

However, the researchers found that when they let the algorithm process all the data, instead of narrowing it down to a specific cancer, the predictions achieved a much higher accuracy. John McDonald, a researcher in this study and the director of Georgia Tech’s Integrated Cancer Research Centre, explains that this is because “on a molecular level, some breast cancers, for example, are going to be more similar to some ovarian cancers than to other breast cancers”. Thus, taking in the data of diverse cancer types provided more accurate predictions of the optimal treatment. Testing this algorithm on 273 ovarian cancer patients, the algorithm successfully predicted their response to 7 of the most commonly prescribed chemotherapeutic drugs with 85% accuracy. This is an impressive achievement considering its innovative take at tackling the ongoing battle of finding a cure to cancer, the first of its kind. Furthermore, its availability as an open source algorithm means that it is readily available for others to use. For the first time, these findings can be utilized by medical institutions, for example. When they input their patients’ profiles, they will have immediate access to predictions for the optimal drug treatments tailored to their cancer. This open source also creates opportunities for collaboration with other cancer institutions. Again, as we think about learning math, the more knowledge we grasp, the better we are able to tackle difficult problems. Similarly, testing this algorithm with larger datasets and even more cancer cases will “teach” the algorithm to understand the genetic profiles.

Thus, it is able to achieve an even higher accuracy when faced with patients of unique cancer conditions. With the success of developing this unconventional approach to cancer treatment, the future of this study is definitely one to look out for. A springboard for greater collaboration, it holds so much potential to advance our progress in cancer research and even tackle cancer once and for all.